SC21 Proceedings

The International Conference for High Performance Computing, Networking, Storage, and Analysis

Regulating Traffic in a Crowded Cache: Overcoming the Container Explosion Problem

Student: Kevin Gao (University of California, Berkeley)
Supervisor: Kyle Chard (Argonne National Laboratory (ANL))

Abstract: Multi-user interactive computing services, such as Binder, dynamically create and deploy software containers to provide customized execution environments with required system and language dependencies. Unfortunately, container creation can be slow, making services unresponsive to users, while caching leads to combinatorial explosion in the number of containers due to the infinite possible combinations, versions, code changes, and users. We analyze 13,946,918 Binder launches and explore caching strategies that consider various features, including package popularity, version stability, recent use, and install time and size. We show that our methods can reduce total storage consumption by 1-3% and creation time by 6-11% when compared with a least recently used strategy.

ACM-SRC Semi-Finalist: no

Poster: PDF
Poster Summary: PDF

Back to Poster Archive Listing